September 2, 2026
Google is rolling out a tool called Google Pics that lets you generate finished graphics -- social posts, flyers, presentation slides -- by typing a prompt instead of dragging shapes around a canvas, according to TechCrunch. That puts Google in direct competition with Canva and Adobe, two companies that built their businesses on the assumption that design tools should feel like design tools: layers, templates, drag handles. Google is betting that assumption is about to expire.
I think Google is right about the direction but not necessarily the speed. Plenty of small-business owners and marketers don't want to learn Canva's interface -- they want a poster that says the right thing and looks professional, and they don't care how it got made. A prompt-first tool removes the learning curve entirely, which is exactly the kind of shift that hurts incumbents who monetize complexity through templates and premium tiers. But design work often needs iteration -- 'move that logo left, make the headline bigger' -- and it's an open question whether prompting is actually faster than dragging once you're past the first draft. Canva has spent years building brand kits, team libraries, and approval workflows that a single generation step doesn't replace yet. For businesses already running marketing campaigns through a broader operations platform, the more interesting trend isn't Google Pics specifically -- it's that AI-first tools are increasingly expected to sit inside the workflow rather than as a separate app you have to context-switch into.
Sequoia-incubated Empirik came out of stealth with $21 million and a pitch that TechCrunch summarized as doing for IT infrastructure what Cursor did for software engineering -- except instead of writing code faster, it's trying to predict outages before they happen. That's a meaningfully different ambition than most of the 'AI agent' funding we've seen this year, which has largely chased developer productivity or customer support deflection.
What I find notable here is the shift from AI that reacts to AI that anticipates. Most IT tooling today tells you something broke after it broke, and the incident response playbook kicks in from there -- which is exactly why we built out a whole incident response section of our own trust practices, because reacting well still matters enormously. Empirik is betting that with enough signal, you can move the intervention earlier. If that works at scale, it's a genuine category shift, not just a faster version of an existing tool. But 'predict outages before they happen' is also one of the oldest promises in enterprise software -- APM vendors have chased this for over a decade with mixed results, because failures are often novel, not repeats of a known pattern. I'd want to see Empirik's false-positive rate before getting too excited; a tool that cries wolf constantly gets ignored, and an ignored prediction system is worse than no system at all. Still, $21 million from a Sequoia incubation is a real signal that sophisticated investors think the pattern-recognition problem in infrastructure monitoring is finally tractable.
Amazon's new Alexa feature, called 'Update Me When,' will send personalized alerts about product launches, tours, books, and shows that might tempt you to buy something, per TechCrunch. Read charitably, it's a helpful notification layer. Read less charitably, it's Amazon training its assistant to proactively manufacture desire on your behalf, using the same AI infrastructure that's supposed to make your life easier.
That tension -- assistant or salesperson -- is going to define a lot of consumer AI over the next year, and it's worth business readers paying attention to even outside retail, because the same proactive-nudge pattern is coming for workplace tools too. The difference between a helpful automated alert and an annoying manipulative one is entirely about whose interest it's serving, and that's a design choice, not a technical inevitability. It's the same reason we think about workflow automation as something that should serve the operator's actual goals, not just maximize engagement with the tool itself.
Which of these actually changes how you work day to day: a prompt-based design tool that skips Canva entirely, or infrastructure software that tries to warn you before something breaks?
Sources